Despite the success of reinforcement learning methods, they have yet to have\ntheir breakthrough moment when applied to a broad range of robotic manipulation\ntasks. This is partly due to the fact that reinforcement learning algorithms\nare notoriously difficult and time consuming to train, which is exacerbated\nwhen training from images rather than full-state inputs. As humans perform\nmanipulation tasks, our eyes closely monitor every step of the process with our\ngaze focusing sequentially on the objects being manipulated. With this in mind,\nwe present our Attention-driven Robotic Manipulation (ARM) algorithm, which is\na general manipulation algorithm that can be applied to a range of\nsparse-rewarded tasks, given only a small number of demonstrations. ARM splits\nthe complex task of manipulation into a 3 stage pipeline: (1) a Q-attention\nagent extracts relevant pixel locations from RGB and point cloud inputs, (2) a\nnext-best pose agent that accepts crops from the Q-attention agent and outputs\nposes, and (3) a control agent that takes the goal pose and outputs joint\nactions. We show that current learning algorithms fail on a range of RLBench\ntasks, whilst ARM is successful.\n